王召德
王召德
> I see that you have added the support for TinyLlama conversion. This is exactly the way I did and got it working. My multilingual model is trained right now....
> > > I see that you have added the support for TinyLlama conversion. This is exactly the way I did and got it working. My multilingual model is trained...
看起来是ONNX的问题，onnx使用1.12.0版本试试看
具体的作用可以看下代码；如果像转换合并的模型就使用`--export`
不同模型适配的情况不一样；一般的语言模型只需要新增一个映射关系的map即可；有些模型结果与原始模型不一样的可能需要增加一些逻辑； 对于多模态模型，需要参考`utils/vision.py`和`utils/audio.py`的实现，添加vision encoder和audio encoder的导出逻辑，必要时还要修改MNN中推理的C++逻辑。
@Juude 可以考虑增加一个开放列表 允许其他用户的modelscope/huggingface等模型链接加入其中，展示在下载列表里
I have read the CLA Document and I sign the CLA
> Hi @wangzhaode, thanks for the PR. It's a great idea and great work! I just tested the PR, and the performances seem interesting. This export is only for half-precision...
All CI checks are passing and the branch is up-to-date with the main branch.